I build systems. I test ideas. I explain what I learn.
I write about architecture, AI-assisted engineering and what I learn while building real systems. When something works, fails or changes how I think, I try to explain why.
THOUGHT
Determinism over probability
Why LLMs can seem so dumb to experienced developers
I kept blaming the model when it made decisions that looked obviously wrong. The useful breakthrough was asking what I knew that the model did not, and which parts of engineering should never have depended on fresh model judgement anyway.
Read the thoughtUse probability for reasoning. Use deterministic software for repeatable truth.
CONCEPT
Agent Learning Loop
An agent hits a problem, repairs it, captures what happened and keeps the useful lesson. Later, another execution can reuse that learning instead of rediscovering the same answer.
Explore the concept- RepairDiagnose the issue, fix it and try again.
- CaptureRecord what failed, why it failed and how the repair was verified.
- PromoteTurn repeatable, useful learning into durable knowledge.
- ReuseRetrieve the lesson in a later execution instead of starting from scratch.
WHY THIS SITE
I am interested in the work behind the answer.
Most of the useful lessons come from the messy bit: the wrong assumption, the failed check, the repair, the decision that should have been written down, or the moment a vague idea becomes something you can actually test.
robbie.build is where I keep those thoughts, concepts, tutorials and projects in a form another person can understand and use.